Application performance and full-stack observability on OpenTelemetry
Send traces, logs and metrics to Rapitt over OpenTelemetry and see each application next to the hosts, tunnels and networks it runs on - one console, no second tool to join by hand.
Applications and infrastructure on one screen
An application that is slow is rarely only an application problem. A database host is short of memory, a tunnel is flapping, a disk is full. Most teams look at the traces in one tool and the hosts in another and join them by hand.
Rapitt keeps both in the same console. A service’s rate, errors and latency sit beside the CPU, memory, disk and network of the hosts it runs on, and a host’s drawer has an Application tab with its slowest traces and error logs.
What you can see
Everything below is built on traces, logs and metrics sent over OpenTelemetry.
Services
Every service with its rate, errors and latency, from rollups that count every span, with sparklines that leave a gap rather than a false zero.
Trace waterfalls
Open a trace as a waterfall with the critical path marked. A span whose parent is not stored is shown as the top of its own branch, not dropped.
Logs beside traces
The logs written during a trace sit next to it. Search logs across services and jump from an error line to its trace.
Business transactions
Define a transaction by service and operation name. See its success rate, percentiles, failure reasons and example traces.
Service and dependency maps
Drawn from the spans: who calls whom, databases, queues and external calls as typed nodes, coloured by error rate and by latency against the service’s own previous day.
Application and infrastructure correlation
Overlay a service’s latency and errors on its hosts’ CPU, memory, disk and network. Rapitt names a host metric that moved with it and says plainly that this is correlation, not cause.
Open standards, one managed agent
Rapitt ingests OTLP over HTTP at your workspace address. Create an ingest token, give your exporter the address and the token, and spans, logs and metrics start arriving. Resources are tied to monitored hosts by a host identifier, so an application always lands under its server.
For hosts that do not already run a Collector, Rapitt manages one: a single command installs an OpenTelemetry Collector that collects host metrics, log files (IIS included), the Windows event log or the journal, and receives OTLP from the applications on the machine. The configuration is generated by your workspace, version-checked and validated before it is applied.
Auto-instrumentation guides
Per-language guides generate the settings for you and include a check that sends one test span:
- Java: the OpenTelemetry Java agent, as a service drop-in or environment
- .NET: auto-instrumentation, with a module for IIS app pools and Windows services
- Python: the distro and
opentelemetry-instrument - Node.js: auto-instrumentations through
NODE_OPTIONS - PHP: the OpenTelemetry extension and auto packages, for PHP-FPM
Honest about what it knows
A trace sampled out is a trace you cannot open, so the console says so: transaction figures are computed over stored spans and are marked as thinned by sampling, while the service figures count everything.
Alerts on a service’s error rate or latency can be judged against a fixed limit or against that service’s own learned normal. Too few requests in the window changes nothing, instead of raising an alarm on three samples.
Where it leads next
- Digital experience: the browser and phone side of the same traces
- AIOps: root-cause analysis that cites the traces and logs it used
- Service management: SLAs and business services over the same services
- API and integrations: send telemetry in, read it back out
Observability questions
Do I have to change my code?
For Java, .NET, Python, Node.js and PHP the guides use OpenTelemetry auto-instrumentation, which attaches to the runtime without code changes. For anything else, use the OpenTelemetry SDK for that language and point its exporter at your workspace.
Which standards does Rapitt use?
OpenTelemetry. Any exporter or Collector that speaks OTLP over HTTP can send traces, logs and metrics, in protobuf or JSON, with an ingest token you create in the console.
Is sampling hiding errors from the service figures?
No. Request, error and latency figures count every span that arrives, even when only some traces are kept. Error spans are always kept.
Is my telemetry mixed with other customers’ data?
No. Each customer has its own isolated database, and telemetry is stored in it.
See it on your own systems
A walkthrough runs on your own estate. Reserve your workspace address while you book.